Uploaded image for project: 'Beam'
  1. Beam
  2. BEAM-8189

Python DataflowRunner fails when using a Shared VPC from another project



    • Type: Bug
    • Status: Closed
    • Priority: Major
    • Resolution: Fixed
    • Affects Version/s: 2.15.0
    • Fix Version/s: 2.16.0
    • Component/s: runner-dataflow
    • Labels:


      The Setup:

      I have two Projects on the Google Cloud Platform

      1) Service Project for my Dataflow jobs

      2) Host Project for Shared VPC & Subnetworks

      The Host Project has configured Firewall Rules for the Dataflow job. ie. allow all traffic, allow all internal traffic, allow all traffic tagged with 'dataflow' etc


      The Args

      --project <host project name>
      --network <shared vpc project name>
      --subnetwork "https://www.googleapis.com/compute/v1/projects/<shared vpc project name>/regions/<region job is running in service project>/subnetworks/<name of subnetwork in shared vpc project>"
      --service_account_email=<service account with Compute Network User permission for both projects, shared vpc network & subnetwork>

      The Problem

      The job will hang when performing shuffle operations. I will also see the following warning:

      The network miles-qa-vpc doesn't have rules that open TCP ports 1-65535 for internal connection with other VMs. Only rules with a target tag 'dataflow' or empty target tags set apply. If you don't specify such a rule, any pipeline with more than one worker that shuffles data will hang. Causes: No firewall rules associated with your network.


      What I've Tried


      1. Only passing "subnetwork" arg without "network" but that only modifies the warning to state "default" instead of "miles-qa-vpc", which sounds like a logging error to me.

      2. Firewall rules have been configured to:

      • allow all traffic
      • allow all internal traffic
      • allow all traffic with the source tag 'dataflow'
      • allow all traffic with the target tag 'dataflow'

      3. Service Account has been configured to have Compute Network User permissions in both projects.

      4. Ensured subnetwork is in the same region as the job.

      5. Network in the service project is happily serving a dedicated cluster for other purposes in the host project.

      It genuinely seems like the spawned Compute Instances are not gaining the configuration.

      I expect the Dataflow job not to report the firewall issue and successfully deal with shuffling (GroupBys etc.)




            • Assignee:
              medwards Miles Edwards
            • Votes:
              1 Vote for this issue
              2 Start watching this issue


              • Created: